The Literature Database is a result of our research and development efforts in AI security. It collects literature related to AI security, including academic papers and blogs published on the internet and features an automatic labeling system for efficient organization. The figure below provides an overview of the Literature Database. This article briefly introduces its key features.
Figure 1. Overview of the Literature Database.
The literature database offers the following three main functions:
- Collection of literature related to AI security
- Content-based labeling of literature
- Automatic posting of labeling information
We launched the Literature Database in March 2025 and are continuously engaged in research and development to address the following technical challenges:
- Accuracy of assigned labels
Since labels are assigned using large language models (LLMs), they are not always guaranteed to be academically accurate. We plan to further enhance the accuracy of the labeling process. - Spelling variations
Variations in notation, such as differences in English expressions and abbreviations, pose a challenge. A key focus moving forward is to develop methods to unify these variations.
This article has introduced the automatic labeling feature utilized in the Literature Database. As the database is based on ongoing research and development, there are many challenges to overcome. We remain committed to improving the Literature Database continually.